Community아트 & 디자인github.com

LevNas/ccaudit

Record and preserve audit trails of AI-assisted Claude Code sessions: auto-flush to a remote trail repository on session end, plus /audit-flush for manual inspection.

지원 대상Claude Code~Codex CLI~Cursor
npx skills add LevNas/ccaudit

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

문서

audit-flush

Manual counterpart to the Stop hook. Same logic, invoked on demand.

When to use

  • The user asks to inspect staged trail entries before they auto-flush.
  • The user wants to retry entries that moved to $CCAUDIT_PENDING_DIR after an earlier failure (secret hit or push error).
  • Auto-flush is disabled (CCAUDIT_MANUAL=1) and entries must be flushed explicitly.
  • The user needs a written record that " inspected and approved this flush at time X" — running /audit-flush leaves a trace (the commit author and timestamp) that an unattended Stop hook cannot.

Flags

FlagBehavior
(none)Same flow as hooks/auto-flush.sh: scan, secret-check, push, clear staging.
--dry-runList what would be flushed, exit without touching anything.
--forceInclude $CCAUDIT_PENDING_DIR contents in addition to staging; retry everything.

Procedure

Follow skills/audit-flush/procedure.md step by step. Do not invent additional safety checks — the hook and this skill must stay behavior-identical, so divergence creates undefined state.

What this skill does NOT do

  • It does not modify trail entries. Entries are append-only once written to staging.
  • It does not delete from the remote trail repository. If a bad entry was pushed, the correction is a new entry referring to the bad one by hash — never a rewrite.
  • It does not reconfigure the remote. CCAUDIT_TRAIL_REPO is read-only from this skill's perspective.

관련 스킬